Patient turnover and nursing staff adequacy

Health Services Research, April, 2006 by Lynn Y. Unruh, Myron D. Fottler

We acquired the ALOS and nursing staff data for each general, acute care Pennsylvania hospital (from 1994 to 2001) from the Pennsylvania Department of Health, patient severity from the Arias MediQual system, and APDC from the American Hospital Association, annual survey of hospitals. Because of hospital openings, mergers, closings, missing data, and outliers, the number of hospitals in the analyses varied from year to year, ranging from 162 to 205.

Data Preparation and Analyses

Using statistical analysis software (SAS), we ran descriptive statistics for the measures as reported in Table 1. To roughly test whether an inverse relationship exists between length of stay and nursing intensity, we plotted existing RN staffing against existing ALOS, using the 8 years of data. The relationship is generally inverse (the plot is not presented here).

As the specific inverse relationship is not known, we developed two functional forms of the inverse to test as indicators. The first form is the full inverse (1/LOS). The second form, the square root of the inverse, is designed to moderate the full effect of the inverse as discussed previously.

We explored two methods of transforming the indicators into indices. The first was to divide the hospital's yearly mean by the 1994 aggregate average. The second was to divide the hospital's yearly mean by the hospital's own 1994 value. As this study examines the impact of patient turnover on staffing measures over time, we chose the latter method. To ascertain differences across hospitals, cross-sectional comparisons could easily employ the first method.

Descriptive statistics for the indices are reported in Table 1. Turnover indices 1 and 2 show a 29 and 13 percent increase, respectively, over the 8-year period. The most significant increases in patient turnover were in the 1995-1998 period. The severity index increased 16 percent overall. Its growth peaked in 1998 and 1999.

Following data preparation, we ran simple fixed effects regression analyses on the turnover indicators to examine their fit to RN staffing. The model was RN/APDC = [[beta].sub.0] [[beta].sup.*.sub.1] turnover indicator [[delta].sub.i]FE [epsilon]. Turnover indicator 1 was significantly related to RN staffing at p < .0001 (coefficient = .00302), and turnover indicator 2 was significant at p < .01 (coefficient = .001823).

Next, we assessed the effect of the indicators on nursing staff measures by constructing a baseline measure composed of the ratio of RNs to standard APDC, and new measures of the ratio of RN to adjusted APDC (APDC multiplied times the patient turnover indices, and times both patient severity and the turnover indices). Finally, we assessed the significance of the difference between the original and new measures of RN staffing by conducting a paired sample t-test of the mean difference in the measures, and in the percent change in measures, in each hospital in each year 1994-2001.

RESULTS

Table 2 presents Pennsylvania hospital RN staffing ratios from 1994 to 2001 before and after patient turnover and severity adjustments, the mean differences between the unadjusted and adjusted measures, and the statistical significance of the differences (t-values). Unadjusted ratios increased until 1996, then decreased slowly thereafter, ending slightly lower in 2001 compared with 1994. Adjusted ratios fell sharply from the beginning. The adjusted staffing ratios were significantly lower than the comparable unadjusted staffing ratios starting in 1995, and the difference became increasingly larger and more significant over time. By the year 2001, the t-values of the mean differences were large and significant at the .0001 level. In real terms, the unadjusted RN staffing ratio of 2.81 in 2001 would be reduced to between 2.09 and 2.56 depending on which of the four patient turnover adjustments is used. This means that the unadjusted 2001 measure overstates the actual RN staffing ratio anywhere from 9 to 26 percent.